AI: More Than a Buzzword

Healthcare

Kelly Gaither
TACC Deputy Director and Director of Visualization

Nothing is more important than the health and well-being of our families, friends, neighbors, and colleagues. Amidst what might seem to be daily conflict in our public discourse, improving health and the access to and delivery of healthcare may be one area in which we can all agree.

We are seeing a re-emergence of well-powered AI methods and tools that are catalyzing new and exciting advances in health and medicine. Large language models are being used to improve the content, quality, and assessment of clinical notes. Ambient AI scribes using machine learning are both alleviating the burden of clinical documentation, a significant factor in physician burnout, and assisting physicians by recommending more empathetic language. Clinical decision support tools are being built with the capability to alert physicians to gaps in patient care. 

AI has been instrumental in providing access to healthcare through telemedicine and AI-assisted chatbots. Machine learning has been used extensively to analyze infection rates in hospitals to personalizing drug delivery for treating cancer to illuminating the most vulnerable members of our population, shining a light on inequities in both access and delivery of care. Exciting advances are being made through the application of deep learning neural networks applied to fetal health, reframing standard of care definitions of gestational age.

The capacity and capability TACC provides to AI methods and tools are crucial for powering new discoveries.

While certainly promising, the long-term impact of AI-enabled rapid advances in health and medicine remains to be seen. Much depends on careful examination and adherence to the ethical practice of  “do no harm.”

We are at the forefront of a new frontier in medicine, one that is embracing AI and computation. 

There is little doubt that we have entered an exciting technological era in which AI is serving as the springboard for new advances aimed at revolutionizing how health and medicine are researched, delivered, and accessed.

Life Sciences

John Fonner
Director, Special Projects

The Life Sciences domain is fertile ground for AI training and inference. 

The study of biology is centered around capturing data of our lived experiences and the underlying physiology that, in complex and nuanced ways, drives those experiences. It is a grand challenge to understand biology at a cellular level (~100 trillion times larger) and to then quantify how cellular processes affect our health as a whole person (another factor of ~30 trillion). 

Through the sustained effort of thousands of researchers, the past 20 years have given society an incredible set of tools and techniques to study life. 

Molecular dynamics simulations give atomistic insight to the behavior and interaction of proteins, viruses, and other biomolecules at scales of hundreds of millions of atoms. This was used to understand how COVID-19 infects cells and
to assess how vaccines interrupt this process. Genome sequencing and the subsequent computing pipelines that turn billions of raw sequence fragments into knowledge about genomic mutations, metabolism, and gene regulation provide a deep statistical foundation connecting physiology and pathology to genetic and cellular factors. These tools are enabling precision medicine that, for example, lets doctors customize the treatment of specific breast cancers based on their underlying genetic causes. 

These breakthroughs have been achieved without using AI as it is currently defined, but the large, rich datasets created by these techniques are already being used to train AI models. As an example, Alphafold 3 was released this year, which replaces the dauntingly complex physics of protein folding with an AI model that generates 3D structures based on input protein sequences. The computing requirements for this model are not trivial, and the results are not perfect, but AI provides a leap forward for problems that are currently intractable with pure physics approaches. 

This ability—using an AI model trained on large data to predict the results of a terribly complex process—maps to much of life sciences research. 

Genotype to phenotype relationships, clinical health outcomes based on treatments, and virtual screening for new therapeutics are all intensely complex and difficult to predict. But because researchers have been diligently collecting data around these areas of study, AI will become a powerful tool to assist with predictions and biological understanding.

Human-AI Interaction

Suzanne Pierce
Research Scientist, Decision Support Systems

AI and computing capabilities are accelerating the ways that researchers integrate and implement solutions for difficult problems. 

Computers are our best partners for solving complex and multi-disciplinary challenges, particularly when there are real-world impacts like those found in disaster management and natural hazards. 

Adapting AI approaches with science-based data and models is the core for next-generation intelligent decision support applications that leverage Human-AI interactions. This includes interfaces that help decision-makers see meaning in data when responding to extreme events; finding solutions to better protect vulnerable communities; or prioritizing how we manage valued natural resources.

At the same time, our own human experience and ability to make sense of what matters is indispensable for ensuring better outcomes that are adapted to the needs of communities. Discussion around the applied uses of AI in more dynamic settings is expanding to include researchers beyond STEM fields, such as social sciences, humanities, and communication researchers. 

Of particular interest are Human-AI partnerships that can accelerate completion of tasks and assist with complex reasoning that exceed a human’s ability to process information. 

Researchers working on these problems may find valuable contributions from AI subfields via approaches in knowledge representation and reasoning, automated planning and scheduling, optimization techniques for combinatorial problems, and advanced sensing capabilities from intelligent robotics. 

Most importantly, however, Human-AI partnerships promise to improve our ability to respond to societal challenges. AI assistants can accelerate community engaged processes and integration of workflows so we can focus on convergent solutions.

Researchers expanding the use of advanced computing and AI tools to improve understanding of science-based decisions will be the most exciting space to watch in the coming years.